Rosace: a robust deep mutational scanning analysis framework employing position and mean-variance shrinkage

Abstract

Deep mutational scanning (DMS) enables functional insight into protein mutations with multiplexed measurements of thousands of genetic variants in a protein simultaneously. The small sample size of DMS renders classical statistical methods ineffective, for example, p-values cannot be correctly calibrated when treating variants independently. We proposeRosace, a Bayesian framework for analyzing growth-based deep mutational scanning data.Rosaceleverages amino acid position information to increase power and control the false discovery rate by sharing information across parameters via shrinkage. To benchmarkRosaceagainst existing methods, we developedRosette, a simulation framework that simulates the distributional properties of DMS. Further, we show thatRosaceis robust to the violation of model assumptions and is more powerful than existing tools underRosettesimulation and real data.

MoCHI: neural networks to fit interpretable models and quantify energies, energetic couplings, epistasis and allostery from deep mutational scanning data

“…The only configuration information strictly required to run MoCHI is a plain text model design file that defines the neural network architecture, and which additionally includes a path to the pre-processed DMS data for each observed phenotype (table rows) as provided by tools such as Enrich2 33 , DiMSum 34 , mutscan 35 or Rosace 36 (see Methods). MoCHI conveniently handles all low-level data manipulation tasks required for model fitting including the definition of training-test-validation data splits and 1-hot encoding of sequence features from AA sequences.…”

“…MoCHI performs model inference accounting for empirical noise (σ n ) in observed phenotype estimates ( y n ) as supplied by the user and provided by tools such as Enrich2 33 , DiMSum 34 , mutscan 35 or Rosace 36 . MoCHI can be configured to train the parameters of genotype-phenotype models assuming a Gaussian noise model: where is the predicted phenotype score of variant n .…”

“…) as supplied by the user and provided by tools such as Enrich2 33 , DiMSum34 , 𝑦 𝑛 mutscan35 or Rosace36 . MoCHI can be configured to train the parameters of genotype-phenotype models assuming a Gaussian noise model:…”

Molecular basis of proton-sensing by G protein-coupled receptors

“…Sequencing files were obtained from the sequencing core as fastq.gz after demultiplexing. The experiment was processed using a DMS-specific pipeline we have developed 63 . The pipeline implemented the following steps: first, adapter sequences and contaminants were removed using BBDuk, then paired reads were error corrected with BBMerge and mapped to the reference sequence using BBMap with 15-mers (all from BBTools 64 ).…”